Hi Brian: The Hop by Hop is certainly the clean solution.
The trouble is that it requires additional bytes in every packet for the header and for the IP-in-IP encapsulation that goes with it; yet RPL operates in a domain where devices can be strictly constrained in energy and frames can be very small, so every bit counts quite dramatically. Thus we're looking for more affordable alternates. We have started a 6MAN document on the IP in IP avoidance for instance. Here, the point is not to get the flow label definition impacted by RPL but to see if the flow label can be used by RPL within the definition that 6MAN will refine. If we have enough mutable bits, then we can allow an alternate mode in RPL whereby those they are used as opposed to the HbH. For the instance ID, it's certainly a lot more complex since we are defining an interaction between the app layer and the network. I do not know if the result of a 6MAN work on such interaction can be used by RPL in the end, but I think it is a good subject to address before we make all bits mutable. My initial take would be that there should at least be one bit somewhere that indicates that the source does not want (a piece of) the FL to be modified. Pascal > -----Original Message----- > From: [email protected] [mailto:[email protected]] On Behalf Of > Brian E Carpenter > Sent: Tuesday, August 10, 2010 11:53 PM > To: Fred Baker (fred) > Cc: Carsten Bormann; [email protected]; [email protected]; Michael Richardson > Subject: ROLL choice to not use the Flow Label > > Thanks for all the enlightment about ROLL. > > My personal conclusion is that the ROLL considerations are too complex and > too subtle to be compatible with using a general-purpose IPv6 header field (i.e. > the flow label) for ROLL purposes. They seem to be an extreme case of the > challenges of defining a local-use regime for the flow label, which have already > been a stumbling block for the various versions of draft-carpenter-6man-flow- > update. > > So IMHO, ROLL/RPL was quite wise to drop the IPv6 header flow label proposal. > > Brian > > On 2010-08-11 08:53, Fred Baker wrote: > > On Aug 10, 2010, at 11:21 AM, Carsten Bormann wrote: > > > >> OK, I'm not talking of "host" as in originates or terminates traffic, but "host" > in the sense of "does not participate in routing". > >> It appears there is no such thing inside a RPL world then. > > > > A RPL or Manet world doesn't have the 1970's-mentality limitation that hosts > have to be stupid. Of all networking architectures, the Internet is the only one > with that delusion. > > > > http://www.ipinc.net/IPv4.GIF > > > > > -------------------------------------------------------------------- > IETF IPv6 working group mailing list > [email protected] > Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 > -------------------------------------------------------------------- -------------------------------------------------------------------- IETF IPv6 working group mailing list [email protected] Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
